The RC Type Effects

In this section, the effects of RC types (SRC vs. ORC) and of RC complex sentence types (SS, SO, OS & OO) were reported and discussed.

To answer the question of how the two RC types differ in acquisition, we examined the data of the RC types by combining both comprehension and production

99

data together. A two-way ANOVA (2 RC Types x 6 Age Groups) with repeated measures presented a significant main effect of RC Type (F(1, 102) = 117.238, p

< .001) and a significant interaction (F(5, 102) = 2.475, p < .05). In order to obtain the simple effects of RC Type in each age group, independent T-tests were adopted.

Results suggested that for each age group SRCs were significantly easier to process than ORCs (3y: t(34) = 2.855, p < .01; 4y: t(20.699) = 3.530, p < .01; 5y: t(34) = 2.986, p < .01; 6y: t(23.607) = 3.234, p < .01; 7y: t(34) = 3.613, p < .01; adults: t(34)

= 4.062, p < .001). Figure 3.2 provides visual illustration.

Figure 3.2 The Mean Scores and Standard Deviations of RCs of each RC Type by Age Group

As to the age effect in each RC type, one-way ANOVAs showed significant differences in each RC type (SRC: F(5, 102) = 30.957, p < .001; ORC: F(5, 102) = 55.841, p < .001). The post hoc comparisons further indicated that in response to SRCs or ORCs, the 6-year-olds gave significantly more accurate responses than the 3-year-olds; the 7-year-olds shot significantly better score than the 3-, 4- and 5-year-olds; the adults got significantly higher mean scores than all of the children. The

100

results implied that children from 4 to 7 years old rapidly acquire both SRCs and ORCs. Nevertheless, children of 7 years old still cannot manage both RC types as well as adults do.

Regarding the effects of RC complex sentence types, we examined the data of the four RC complex sentence types by combining both comprehension and production data together. A two-way ANOVA (4 RC Complex Sentence Types x 6 Age Groups) with repeated measures reported a significant main effect of RC Complex Sentence Type (F(3, 306) = 70.372, p < .001) but no significant interaction (F(15, 306) = 1.440, p > .05). The post hoc comparisons suggested that the accuracy of SS (M = 0.57, SD = 0.22), SO (M = 0.52, SD = 0.17) and OS (M = 0.58, SD = 0.22) was significantly higher than that of OO (M = 0.39, SD = 0.18). When the accuracy data from each group was entered into ANOVAs, significance was reported. The post hoc comparisons of the age groups suggested that for the children aged from 3 to 5 SS, SO and OS were significantly easier than OO, while for the children aged 6 and 7 years old as well as the adults SS and OS were significantly easier than OO. From Figure 3.3 we can observe that numerically speaking, for the 3- and 4-year-olds the accuracy of SS, SO and OS was very close, yet for the 4-year-olds the accuracy of SS and OS became slightly higher than SO. The gap between SS/OS and SO enlarged when children reached 6 years old; therefore, it is observed that for the children aged 6 or more as well as the adults the ranking hierarchy of the four RC complex sentence types in descending order of mean score seems to be SS = OS > SO > OO.

101

Figure 3.3 The Mean Scores and Standard Deviations of RCs of each RC Complex Sentence Type by Age Group

As to the age effect in each RC complex sentence type, one-way ANOVAs showed significant differences in each RC complex sentence type (SS: F(5, 102) = 24.606, p < .001; SO: F(5, 102) = 27.810, p < .001; OS: F(5, 102) = 25.379, p < .001;

F(5, 102) = 37.435, p < .001). The post hoc comparisons indicated that in response to SS, the 6- and 7-year-olds gave significantly more accurate responses than the 3-year-olds; the adults shot significantly better score than all of the children. In responses to SO, the adults got significantly higher mean scores than all of the children. In responses to OS, the accuracy of the 6-year-olds was significantly higher than that of the 3-year-olds; the 7-year-olds were significantly better than the children aged from 3 to 5; the mean score of the adult group was significantly better than that of all of the child groups. As for OO, the 6-year-olds got significantly better score than the 3- and 4-year-olds; the 7-year-olds gave significantly more correct responses than the 3-, 4- and 5-year-olds; the adults were significantly better than all of the children. The results implied that the development of SS, OS and OO is steady, while the

102

acquisition of SO seems to be at a standstill. In addition, the children of 7 years old still cannot manage all the four RC complex sentence types as well as adults do.

To summarize, in line with the hierarchical distance hypothesis (O’Grady, 1997), the accessibility hierarchy hypothesis⁶ (Keenan & Comire, 1977, 1979) and the previous experimental studies (Cheng, 1995; Hsu, 2006), the results showed that both of the children and the adults were much better at SRCs than ORCs, a finding which leads us to believe that Mandarin speakers tackle sentences in a hierarchical way regardless of age difference. As O’Grady (1997) pointed out, the shorter the distance between the gap and its filler in a hierarchical structure, the easier the processing of the certain structure. Since the distance between the gap and the RC head in the hierarchical structure is shorter in SRCs, the accuracy of SRCs is then higher than that of ORCs. With respect to the results of the four RC complex sentence types, the statistics indicated that OO bears a lot more processing burden than the other three types. Though significant differences were not found among SS, SO and OS, it is numerically suggested that for Mandarin speakers aged 6 years old or more the ranking hierarchy in descending order of mean score seems to be SS = OS > SO > OO.

This ranking hierarchy is similar to what was found in Wu et al.’s corpus study (2011), in which they proposed that when all verb types were included in the RC constructions, the ranking hierarchy in descending order of frequency was SS > OS >

SO > OO. Taking the findings of RC types and RC complex sentence types together into consideration, we might conclude that the effect of RC types overrides the effect of directionality in that SRCs are always easier to process than ORCs no matter whether they are in left-branching or right-branching positions. Furthermore, the

6 According to the Accessibility Hierarchy, in the formation of RCs, a subject is more accessible than a direct object, which in turn is more accessible than an indirect object, followed by an oblique, a genitive and an objet of comparison. The more accessible a NP is, the easier the NP can be relativized.

103

ranking hierarchy appears to reject the NVN-schema hypothesis (Bever, 1970), which claims that OS, the first NVN of which conforms to the unmarked SVO word order, is the easiest structure among the four RC complex sentence types. Nevertheless, we may not make the conclusion because the NVN-schema hypothesis only accounts for comprehension, while our finding comes from both production and comprehension tasks. The detailed discussion about the hypothesis is presented in Chapter Five.

As to the results relevant to age effects, we discovered that in spite of the RC types and the RC complex sentence types, the 7-year-olds still could not manage them equally well as the adults do. The development of SRC and that of ORC are identical in that the great progress appears when children reach 6 years old. The same development pattern is also tenable in SS, OS and OO, yet the acquisition of SO remains steady across the child groups. The stagnant development of SO may be explained by considering the effect of directionality and the effect of RC types. As shown in Figure 3.3, in the 3-year-old group the accuracy of SO was slightly higher than that of SS, which in turn was a bit higher than that of OS. When children become 4 years old, the minute differences vanished. By the time they were 5 years old, the mean score of SO became lower than those of SS and OS. The imperceptible changes appear to suggest that for the 3-year-olds the basic nature of a language is a more important factor influencing how they deal with a sentence. More specifically, left-branching structures, the essence of Mandarin, play an important role for these young children than older ones; therefore, the mean scores of the RCs in left-branching positions, i.e., SO and SS, were slightly higher than that of the RCs in right-branching positions, i.e., OS. However, when children reached 5 years old, the effect of RC types seemed to replace the effect of directionality so that the accuracy of SRCs in both branching directions, i.e., SS and OS, became slightly higher than that of SO.

104

The effect of RC types kept influencing with the increase of age, making the development of SO lingering.

3.4.3 The Task Effects

To investigate the task effect, the comprehension and production data were examined separately. Two-way ANOVAs (2 Tasks x 6 Age Groups) with repeated measures presented a significant main effect of Tasks (F(1, 102) = 671.155, p < .001) and a significant interaction (F(5, 102) = 36.938, p < .001) in complex sentences with left-branching RCs as well as a significant main effect of Tasks (F(1, 102) = 251.781, p < .001) and a significant interaction (F(5, 102) = 35.195, p < .001) in complex sentences with right-branching RCs. In order to obtain the simple effect of Tasks in each branching direction, we entered the accuracy data from each age group into separate independent T-tests. Results revealed that when RCs were in the left-branching positions, the comprehension accuracy was significantly higher than the production accuracy in each age group (3y: t(27.828) = 24.302, p < .001; 4y: t(34) = 25.683, p < .001; 5y: t(34) = 11.689, p < .001; 6y: t(34) = 8.161, p < .001; 7y:

t(29.770) = 6.591, p < .001), except the adult group (t(34) = -.446, p = .658). When RCs were in the right-branching positions, the same significant differences were found in the children aged from 3 to 6 (3y: t(34) = 25.884, p < .001; 4y: t(34) = 14.377, p < .001; 5y: t(34) = 9.071, p < .001; 6y: t(20.918) = 3.184, p < .01). Though the effects of the two tasks did not reach significance in the 7-year-old group, it is numerically suggested that for the 7-year-olds the accuracy in the comprehension task was also higher than that in the production task (t(23.057) = 1.657, p > .05). However, for the adults, the production accuracy was significantly higher than the comprehension accuracy (t(34) = -.2.338, p < .05). Figures 3.4 and 3.5 present the

105

mean scores and standard deviations of the two tasks by each age group in the left-branching RC construction and in the right-left-branching RC construction, respectively.

Figure 3.4 The Mean Scores and Standard Deviations of the Left-branching RC Comprehension and Production by Age Group

Figure 3.5 The Mean Scores and Standard Deviations of the Right-branching RC Comprehension and Production by Age Group

Regarding the RC development in comprehension and production of different branching directions, one-way ANOVAs reported significant differences among age groups in production of left-branching RCs (F(5, 102) = 72.735, p < .001), comprehension of right-branching RCs (F(5, 102) = 4.109, p < .01) and production of

106

right-branching RCs (F(5, 102) = 53.858, p < .001). The post hoc comparisons demonstrated that the 6-year-olds produced left-branching RCs significantly better than the 3- and 4-year-olds; the 7-year-olds generated significantly more accurate left-branching RCs than the children aged from 3 to 5; the adults got significantly higher score than all of the children. In comprehending right-branching RCs, the adults shot significantly better score than the 6- and 7-year-olds. As to the production of right-branching RCs, the 6- and 7-year-olds were significantly better than the children aged from 3 to 5; the adult group generated significantly more accurate right-branching RCs than all of the child groups. To sum up, the mean scores of comprehension accuracy of left-branching RCs did not differ across age groups, while the mean score of production accuracy of left-branching RCs rise with the increase of age. Similarly, though the mean score of comprehension accuracy of right-branching RCs was significantly lower in the 6- and 7-year-old groups than in the adult group, the differences among the mean scores of the child groups did not reach significance. On the other hand, there was a clear and steady rise in right-branching RC production with the increase of age.

In respect of the Task effect in each RC type, two-way ANOVAs (2 Tasks x 6 Age Groups) with repeated measures presented a significant main effect of Tasks (F(1, 102) = 212.193, p < .001) and a significant interaction (F(5, 102) = 31.170, p < .001) in SRCs as well as a significant main effect of Tasks (F(1, 102) = 735.811, p < .001) and a significant interaction (F(5, 102) = 26.841, p < .001) in ORCs. In order to obtain the simple effect of Tasks in each RC type, we entered the accuracy data from each age group into separate independent T-tests. Results revealed that in the SRC type, the comprehension accuracy was significantly higher than the production accuracy in each child group (3y: t(34) = 20.261, p < .001; 4y: t(34) = 12.550, p

107

< .001; 5y: t(34) = 8.313, p > .001; 6y: t(26.441) = 2.984, p > .01), except the 7-year-old group, in which the trend was only numerically suggestive (7y: t(24.205) = 1.835, p > .05). On the contrary, the adults’ SRC production was significantly better than SRC comprehension (t(24.227) = -3.872, p < .01). In the ORC type, the better RC comprehension than RC production was found in each child group (3y: t(34) = 22.395, p < .001; 4y: t(34) = 37.398, p < .001; 5y: t(34) = 14.910, p < .001; 6y: t(24.645) = 10.460, p < .001; 7y: t(21.989) = 8.268, p < .001), while the same trend was only numerically suggestive in the adult group (t(34) = 1.183, p > .05). Figures 3.6 and 3.7 illustrate the results.

Figure 3.6 The Mean Scores and Standard Deviations of the SRC Comprehension and Production by Age Group

108

Figure 3.7 The Mean Scores and Standard Deviations of the ORC Comprehension and Production by Age Group

Regarding the RC development in comprehension and production of different RC types, one-way ANOVAs reported significant differences among age groups in production of SRCs (F(5, 102) = 43.188, p < .001), comprehension of ORCs (F(5, 102) = 3.509, p < .01) and production of ORCs (F(5, 102) = 55.801, p < .001). The post hoc comparisons demonstrated that the 6- and 7-year-olds produced significantly more accurate SRCs than the children aged from 3 to 5; the adults performed significantly better than all of the children. In comprehending ORCs, the adults shot significantly better score than the 3-year-olds. As to the production of ORCs, the 6-year-olds generated significantly better than the 3- and 4-6-year-olds; the 7-6-year-olds got significantly better score than the children aged from 3 to 5; the adults performed significantly better than all the child groups. To sum up, the mean scores of comprehension accuracy of SRCs did not differ across age groups, yet the mean score of production accuracy of SRCs rise with the increase of age. Likewise, though the mean score of comprehension accuracy of ORCs was significantly higher in the adult group than in the 3-year-old group, the differences among mean scores of the child

109

groups did not reach significance. On the other hand, there was a clear and steady rise in ORC production with the increase of age.

Concerning the Task effect in each RC complex sentence type, two-way ANOVAs (2 Tasks x 6 Age Groups) with repeated measures presented a significant main effect of Tasks (F(1, 102) = 240.775, p < .001) and a significant interaction (F(5, 102) = 28.562, p < .001) in the SS type, a significant main effect of Tasks (F(1, 102)

= 1023.576, p < .001) and a significant interaction (F(5, 102) = 22.774, p < .001) in the SO type , a significant main effect of Tasks (F(1, 102) = 127.688, p < .001) and a significant interaction (F(5, 102) = 21.868, p < .001) in the OS type as well as a significant main effect of Tasks (F(1, 102) = 213.463, p < .001) and a significant interaction (F(5, 102) = 16.510, p < .001) in the OO type. In order to obtain the simple effect of Tasks in each RC complex sentence type, we entered the accuracy data from each age group into separate independent T-tests. Results revealed that in the SS type, the comprehension accuracy was significantly higher than the production accuracy in each child group (3y: t(34) = 16.444, p < .001; 4y: t(30.380) = 14.512, p

< .001; 5y: t(34) = 7.977, p < .001; 6y: t(34) = 4.291, p < .001; 7y: t(34) = 2.230, p

< .05), while the adults produced better than comprehending SS RCs (t(24.531) = -3.489, p < .01). In the SO type, the better RC comprehension than RC production was found in each age group (3y: t(20.183) = 17.332, p < .001; 4y: t(25.323) = 30.051, p

< .001; 5y: t(21.929) = 15.185, p < .001; 6y: t(34) = 11.452, p < .001; 7y: t(34) = 12.728, p < .001; adults: t(34) = 3.140, p < .01). In the OS type, the better RC comprehension than RC production was only discovered in the children aged from 3 to 6 years old (3y: t(34) = 15.136, p < .001; 4y: t(34) = 9.800, p < .001; 5y: t(34) = 7.384, p < .001; 6y: t(24.604) = 2.185, p < .05). Significant differences were not found in the 7-year-old group (t(34) = .973, p > .05), while the adults were found to

110

produced OS RCs significantly better than comprehending the construction (t(34) = -2.898, p < .01). In the OO type, the children shot significantly better score in comprehension than in production (4y: t(34) = 18.554, p < .001; 5y: t(34) = 8.776, p

< .001; 6y: t(17.000) = 5.132, p < .001; 7y: t(19.290) = 2.385, p < .05)⁷, while the adults comprehended and produced OO RCs equally well (t(34) = -.404, p > .05).

Figures 3.8, 3.9, 3.10 and 3.11 illustrate the results.

Figure 3.8 The Mean Scores and Standard Deviations of the SS Comprehension and Production by Age Group

7 The statistics cannot calculate the t-value for the 3-year-old group because the standard deviation of the comprehension task and that of the production task are both zero.

111

Figure 3.9 The Mean Scores and Standard Deviations of the SO Comprehension and Production by Age Group

Figure 3.10 The Mean Scores and Standard Deviations of the OS Comprehension and Production by Age Group

112

Figure 3.11 The Mean Scores and Standard Deviations of the OO Comprehension and Production by Age Group

Regarding the RC development in comprehension and production of different RC complex sentence types, one-way ANOVAs reported significant differences among age groups in production of SS (F(5, 102) = 46.106, p < .001), production of SO (F(5, 102) = 54.424, p < .001), production of OS (F(5, 102) = 35.648, p < .001), comprehension of OO (F(5, 102) = 6.422, p < .001) and production of OO (F(5, 102)

= 33.226, p < .001). The post hoc comparisons demonstrated that the 6-year-olds produced significantly more accurate SS, OS and OO RCs than the 3- and 4-year-olds;

the 7-year-olds shot significantly better score than the children aged from 3 to 5; the adults performed significantly better than all of the children. In producing SO RCs, the adults got significantly better score than all of the children. As to the comprehension of OO RCs, the adults comprehended significantly better than the 3- and 6-years-olds. To sum up, the mean scores of comprehension accuracy of SS, SO and OS did not differ across age groups, yet the mean score of production accuracy of these RC complex sentence types rise with the increase of age. Though the mean score of comprehension accuracy of OO was significantly higher in the adult group

113

than in the 3- and 6-year-old groups, the differences among mean scores of the child groups did not reach significance.

The main finding of this section is that RC comprehension accuracy was significantly higher than production accuracy in the child groups, except right-branching RCs, SRCs and OS RCs in the 7-year-old group, in which the trend was only numerically suggestive. For the adults, better RC comprehension accuracy than production accuracy was only found in SO RCs, while the significant difference was reversed in right-branching RCs, SRCs, SS RCs and OS RCs. The results might purport that the sentence processing of the 7-year-old children is more analogous to that of the adults than that of the younger children are because the difficulty in the 7-year-olds’ producing complex sentences caused by working memory required for planning message as well as mapping the phrasal constituents and the hierarchical structure is not significantly larger than the difficulty in their comprehending the same construction. Stated another way, the working memory capacity of the 7-year-olds may be more similar to that of the adults. In addition, in the adult group, in which the subjects are supposed to have average working memory capacity, the higher mean scores of right-branching RCs, SRCs, SS RCs and OS RCs in production than in comprehension may imply that speakers tend to cope with sentences successively and in a hierarchical way. To tackling sentences successively, a speaker does not plan much structure to the right of the heads of constituents currently being generated but deal with one clause at a time (Gibson, 1998). Right-branching RCs, involving OS RCs, allow a speaker to tackle main clauses first and the RCs later, hence, should be

The main finding of this section is that RC comprehension accuracy was significantly higher than production accuracy in the child groups, except right-branching RCs, SRCs and OS RCs in the 7-year-old group, in which the trend was only numerically suggestive. For the adults, better RC comprehension accuracy than production accuracy was only found in SO RCs, while the significant difference was reversed in right-branching RCs, SRCs, SS RCs and OS RCs. The results might purport that the sentence processing of the 7-year-old children is more analogous to that of the adults than that of the younger children are because the difficulty in the 7-year-olds’ producing complex sentences caused by working memory required for planning message as well as mapping the phrasal constituents and the hierarchical structure is not significantly larger than the difficulty in their comprehending the same construction. Stated another way, the working memory capacity of the 7-year-olds may be more similar to that of the adults. In addition, in the adult group, in which the subjects are supposed to have average working memory capacity, the higher mean scores of right-branching RCs, SRCs, SS RCs and OS RCs in production than in comprehension may imply that speakers tend to cope with sentences successively and in a hierarchical way. To tackling sentences successively, a speaker does not plan much structure to the right of the heads of constituents currently being generated but deal with one clause at a time (Gibson, 1998). Right-branching RCs, involving OS RCs, allow a speaker to tackle main clauses first and the RCs later, hence, should be